1 Introduction
Classical recommendation systems use long-term behavior records of users to build user portraits and recommend contents based on this [1]–[3]. However, long-term behavior records of users are not available in all cases. For example, when a user is browsing a new app, if the app can’t push the content that the user is interested in based on the user’s short-term behavior, the app may lose the user. So the Session-based Recommendation(SBR) problem has received a lot of attention in recent years.